Single image super-resolution via online dictionary learning with double regularization parameters

Abstract

The performance of single image super-resolution (SR) based on sparse coding is promising but the artifacts are obvious. In order to promote the SR efficiency, the proposed algorithm employs Online Dictionary Learning (ODL) to train the overcomplete dictionaries separately to reduce the artifacts. It sets two different regularization parameters in the phases of dictionary learning and image reconstruction. They are tuned independently to get the best results. In the experiments, the PSNRs are 0.41dB higher than Sparse Coding Super-Resolution (SCSR) and 0.28dB higher than ODL SR algorithm in average. The proposed method can eliminate the artifacts and recover the texture details efficiently.